Professional Documents
Culture Documents
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
a r t i c l e
i n f o
Article history:
Received 15 December 2008
Received in revised form 19 May 2009
Accepted 1 June 2009
s u m m a r y
A bathymetric survey of 575 km of the central Amazon River and one of its tributaries, the Purus, are
combined with gauged data to characterise the Amazon ood wave, and for hydraulic modelling of the
main channel for the period June 1995March 1997 with the LISFLOOD-FP and HEC-RAS hydraulic models.
Our investigations show that the Amazon ood wave is subcritical and diffusive in character and, due
to shallow bed slopes, backwater conditions control signicant reach lengths and are present for low and
high water states. Comparison of the different models shows that it is necessary to include at least the
diffusion term in any model, and the RMSE error in predicted water elevation at all cross sections introduced by ignoring the acceleration and advection terms is of the order of 0.020.03 m. The use of a wide
rectangular channel approximation introduces an error of 0.100.15 m on the predicted water levels.
Reducing the bathymetry to a simple bed slope and with mean cross section only, introduces an error
in the order of 0.5 m. These results show that when compared to the mean annual amplitude of the Amazon ood wave of 1112 m, water levels are relatively insensitive to the bathymetry of the channel
model. The implication for remote sensing studies of the central Amazon channel, such as those proposed
with the Surface Water and Ocean Topography mission (SWOT), is that even relatively crude assumptions
regarding the channel bathymetry will be valid in order to derive discharge from water surface slope of
the main channel, as long as the mean channel area is approximately correct.
2009 Elsevier B.V. All rights reserved.
Introduction
The ow of the Amazon River accounts for around 20% of total
global continental runoff (Richey et al., 1989b) and the dynamic
exchange of water between the channel and oodplain play a central role in biological and biogeochemical processes in the Amazon
basin (Junk and Piedade, 1993; Melack and Forsberg, 2001; Wittmann et al., 2004). The monomodal ood pulse that passes annually down the Amazon River is a key driver for this hydrological
system and understanding the behaviour and characteristics of this
ood wave is therefore important to any study attempting to quantify processes dependent upon it.
Hydrological modelling has long been used as a tool to understand and simulate the Amazon River system, including using Mus-
* Corresponding author. Tel.: +44 (0) 117 928 9954; fax: +44 (0) 117 928 7878.
E-mail addresses: mark.trigg@bristol.ac.uk (M.A. Trigg), matthew.wilson@sta.
uwi.edu (M.D. Wilson), paul.bates@bristol.ac.uk (P.D. Bates), horrittm@halcrow.
com (M.S. Horritt), alsdorf@geology.ohio-state.edu (D.E. Alsdorf), brforsberg@gmail.
com.br (B.R. Forsberg).
0022-1694/$ - see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2009.06.004
inundation of the Amazon River and oodplain using 2D hydraulic modelling at this level of detail, using improved digital elevation data for the Amazon basin from the Shuttle Radar
Topography Mission (SRTM) own in 2000 (Rodriguez et al.,
2006).
Hydraulic modelling has the potential to provide a level of
detailed information regarding ood duration and depth for the
Amazon River and oodplain that will enable a host of other research on processes that depend upon these important variables.
An understanding of the basic characteristics of a ood wave and
good bathymetric data are key to the application of appropriate
hydraulic methods to the study of a river systems hydraulics
and the accuracy of any further study based on the results of
these methods. Unfortunately, there are very few studies that
provide information related specically to the hydraulic characteristics of the Amazon ood wave. Of the studies to date the
most signicant for our analysis are those of Meade et al.
(1991) and Alsdorf et al. (2005). Meade et al. (1991) used analysis of the stage discharge behaviour of river gauging stations to
demonstrate the extensive backwater effects along the mainstem. Alsdorf et al. (2005) used linear diffusion modelling to
characterise oodplain ow, showing the water elevation from
the main channel propagates across the oodplain as a diffusion
wave.
In addition to SRTM, other remote sensing data have provided
insights into the variations of surface water height and ood extents of the rivers and wetlands in the basin and so provide a
wealth of calibration and validation datasets for hydraulic modelling. In particular, satellite radar altimetry (Birkett et al., 2002),
passive microwave systems (Hamilton et al., 2002) and Synthetic
Aperture Radar (SAR) data (Alsdorf et al., 2000; Rosenqvist et al.,
2002; Hess et al., 2003; Melack et al., 2004) have all been used to
quantify variations in Amazon River and oodplain surface water
extent and elevation. However, a limitation of the use of many remote sensing methods in hydraulic modelling is the inability of the
optical or radar signal to penetrate the water surface, thus providing little or no bathymetric detail for the main river channels and
oodplain channels. Bonnet et al. (2008) demonstrated it is possible to use a time series of SAR images (Martinez and Le Toan, 2007)
to map periodically ooded areas of oodplain, but the study still
required bathymetric data collected by Acoustic Doppler Current
Proler (ADCP) for the permanently ooded areas such as the
channels.
While some bathymetric data are available for isolated gauging
stations and localised study areas, partly due to the Amazons scale
and the fact that sediment dynamics play an active role in shaping
and reshaping the bed forms (Mertes, 1994; Mertes et al., 1996;
Dunne et al., 1998), there is no consistent up-to-date bathymetry
dataset for the Amazon channel. This lack of data is commonly
tackled by applying simplications such as using channels with
mean widths and slopes (Wilson et al., 2007) or by using old shipping charts and assumptions about bed slope (LeFavour and Alsdorf, 2005; Leon et al., 2006). Satellite missions which seek to
derive discharge from space, such as the proposed Surface Water
and Ocean Topography (SWOT) instrument, must also rely on
assumptions regarding the bathymetry data and this has been
identied as a potentially signicant source of uncertainty for such
undertakings (Durand et al., 2008).
In this paper, we identify the relative importance of these
bathymetric issues for hydraulic modelling and other studies that
relate to the central Amazon River channel. This is achieved in
three stages. Firstly we describe the collection of channel bathymetry data for a 430 km reach of the Amazon, known as the Solimes
at this location, and a 145 km reach of its tributary, the Purus
(Fig. 1). These data provide a level of information of a consistent
standard for analysis that has not been previously available for
93
the main channel and facilitate the second and third stages of
the study. Secondly, we undertake a study of the Amazon channel
ow using parameters derived from the bathymetric data in order
to understand the basic hydraulic characteristics of the Amazon
ood wave. As well as providing key hydraulic properties, this ensures a sound theoretical basis for the correct application of
hydraulic models to the main channel. The nal stage uses the data
and understanding derived from the rst two stages to enable the
construction of detailed 1-dimensional (1D) hydraulic models of
the Amazon main channel and the Purus. Experiments were then
undertaken with these models to see the effect of simplifying the
physical process and channel geometry representation on ood
wave routing by comparing model results to independent water
surface elevation data derived from ground gauging stations and
satellite radar altimetry.
This work helps to address three key questions. Firstly, what are
the basic hydraulic characteristics of the Amazon ood wave? Secondly, what is the simplest physical and geometrical representation of the channel ow that can be used in hydraulic models
and still have predictions of water levels match available observed
data? Thirdly, how important are the changing bed conditions to
the overall hydraulics of the channel ow? Understanding the answers to these questions will enable appropriate assumptions to be
made regarding ow modelling where limited data are available
for the central Amazon channel and oodplain. It will also facilitate
future work, including appropriate scaling up of the modelling
work undertaken by Wilson et al. (2007) to a larger domain size,
as well as improving the channel representation which drives
oodplain dynamics.
Channel bathymetry
Bathymetric data collection and interpolation
Bathymetric data provides essential rst order parameters for
the study of a rivers hydraulics or the application of hydraulic
modelling to a river. In this section we document the bathymetric
data collection done for this study of the central Amazon River
channel and one of its tributaries the Purus. This data collection
was necessary as there were no up-to-date data available at the level of detailed required for this study. Sonar data were collected 8
21 June, 2005, just after the high water peak, using a Lowrance
sonar connected to a GPS unit tted to a small boat. The total distance surveyed was around 1900 km over 14 days, along a 430 km
reach of the Solimes between Manaus and Coari and a 145 km
reach of the Purus beginning at the conuence with the Solimes
and ending just downstream of Arum. A total of 234,863 sonar
data points were collected for the Solimes and Purus reaches.
The survey route was chosen such that diagonal rather than
perpendicular cross sections were obtained, which allowed the
coverage of extensive reaches relatively quickly so that as much
detail of the whole channel as possible was collected rather than
just the representative cross sections normally required for
hydraulic modelling. The aim was to create as complete a three
dimensional channel as possible in order to select later the required cross sections as well as increase potential utility of the
resulting data. The nal collection path is shown in Fig. 2. All
branches of the main channel were surveyed as well as a section
of one of the main oodplain channels.
Sonar depths were converted to bed elevations using a planar
water surface level as a reference, which was tted to stage records
from Itapeu and Manacapur. During the 14 day period of the survey stage elevations changed by 10 cm at Itapeu and 23 cm at
Manacapur. All gauged data were tied to a common elevation datum which uses a geoid model (Kosuth et al., 2006). The resulting
bed elevations were then sampled onto a triangular irregular mesh
94
Fig. 1. (a) Location overview. (b) Study site centred on Solimes and Purus conuence, showing the river channels with background SRTM DEM, and the locations of available
river stage data (Manacapur, Anam, Codajs and Beruri gauging stations) and satellite altimetry data relevant to the river channel used for model validation.
95
Fig. 3. 90 m bathymetry elevation grid, with three insets. Elevations shown as 10 m interval contours for clarity.
For the Solimes reach, the channel width varies from 1620 m
to 5624 m with a mean of 3711 m and the minimum bed elevation
varies from 26.45 m to 8.03 m with a mean of 8.44 m (Kosuth
et al. (2006) datum). This means that signicant portions of the
channel are well below sea level even though the centre of the
study location is approximately 1500 km inland (a third of the
Amazon Rivers total length), which further illustrates the extraordinary nature of the Amazon River and also explains the very shallow water slopes. At high water, channel depths range from 20 m
to 52 m. For the Purus reach the channel width varies from
600 m to 1678 m with a mean of 1114 m and the minimum bed
elevation varies from 9.78 m to 9.48 m with a mean of 2.08 m.
At high water, channel depths range from 16 m to 35 m.
96
Sf S 0
@y u @u 1 @u
@x g @x g @t
where, Sf is the friction slope, S0 the bed slope, y is the depth, x the
distance along the river, u the velocity averaged over a cross section,
t time and g the acceleration due to gravity (9.81 m s2). The two
most common approximations are the kinematic wave approximation which is represented by the rst right hand side term, and the
diffusive wave represented by the rst and second right hand side
terms.
Two methods for characterising the hydraulics of the Amazon
were identied and applied to the Amazon ood wave. The rst
of these was developed by Vieira (1983). This method derives from
an extensive numerical analysis comparing different approximations of the Saint Venant equations applied to a range of hydraulic
conditions. The non-dimensionalised parameter space dened by
F0, the Froude number and k, the kinematic wave number is divided up into zones where the various approximations were found
to be valid. Application of this method is straightforward and involves estimating the two parameters, F0 and k, and identifying
the valid approximation that applies. The formulations of the two
parameters used by Vieira that were applied in our analysis are:
F0
k
gy1=2
S0 L
yF 20
V 20
gy0
TV 0 Sf 0
T
y0
F 20
4
5
Hydraulic characteristics
Henderson (1966) classied a shallow wave as one where the
ratio of water depth to wave length is usually less than 0.05. For
the Amazon ood wave, the depth of ow is between 20 and
30 m, with an annual wave length resulting in a ratio of approximately 3 105. Unsurprisingly, we can therefore see that the
shallow wave description assumption is appropriate.
For application of the characterisation method of Vieira (1983)
and that of Moussa and Bocquillon (1996), estimates of the parameter values for the Amazon channel were made using the bathymetric data described in the preceding section and data from the ANA
Itapeu and Manacapur gauging stations. A summary of the key
parameter values for the Solimes and Purus reaches are presented
in Table 1 and results of the characterisation methods for the Amazon ood wave on the Solimes are presented in Table 2.
Both characterisation methods use plots of the parameter space
divided into zones where different approximations are appropriate. The results in Table 2 are plotted on these characteristic plots
in order to graphically illustrate the results, Figs. 4 and 5. It should
be emphasised that these results are derived from estimates and
assumptions relevant to the central section of the Amazon considered in this paper, but none-the-less provide an important insight
into the overall hydraulic characteristics of the Amazon ood
wave.
The results of both characterisation methods place the Amazon
ood wave in the zone where a diffusive approximation is valid.
Flow in the channel is entirely in the subcritical range. Of particular
interest in these results is the relatively extreme nature of the values derived for the Amazon compared to those more commonly
encountered in typical rivers. This is illustrated by the fact that
we have had to extend both characteristic plots well beyond the
range originally envisaged by their creators as being sufcient for
most ood waves. This limitation of existing methods applied to
the Amazon has also been noted by Latrubesse (2008), who classies the Solimes at this location as an anabranching mega river. In
general terms, the ow is relatively slow moving and deep, resulting in a very low Froude number. Water slopes are extremely mild
and interestingly both the Solimes and Purus water surface slopes
are shallower than their mean bed slopes, particularly for the
Purus, indicating the potential importance of the backwater on
both reaches.
Hydraulic modelling
Model choice and development
With the bathymetric data collected and the results of our ood
wave characterisations, it was possible to undertake a 1D hydrodynamic modelling analysis of the main river channels, without
oodplains and ascertain the relative importance of the bathymetric data and the process representation.
LISFLOOD-FP was chosen for the initial modelling work carried
out on the Amazon by Wilson et al. (2007), since it was developed
to provide rapid calculation for large 2D domains (Bates and De
Roo, 2000). In this paper we further develop this model as part of
an overall aim to understand better the complex river and ood-
Table 1
Summary of key parameter values for Solimes and Purus reaches.
Parameter
Description
Units
V
y
S0
Sf
Mean
Mean
Mean
Mean
m/s
m
cm/km
cm/km
0.77
21.91
4.96
2.92
1.30
33.50
4.96
2.90
0.26
10.93
6.50
1.00
0.86
23.13
6.50
1.66
velocity
depth
bed slope
water Slope
Description
Value
F0
k
F 20
T+
Froude number
Kinematic wave number
Froude number squared
Non-dimensionalised period
0.0524
82.7
0.0027
32.3
Fig. 4. Extended plot after Vieira (1983) showing the parameter values for the
Amazon ood wave.
97
@Q @A
q
@x @t
4=3
n2 P Q 2
@y
0
S0
10=3
@x
A
6
7
where, Q is the volumetric ow rate in the channel, A the cross sectional area of the ow, q the ow into the channel from other
sources, n the Mannings coefcient of friction, and P the wetted
perimeter of the ow. In this case, the channel is assumed to be
wide and shallow, so the wetted perimeter is approximated by
the channel width, which is reasonable for the Amazon reach studies here. In (7) the term in brackets is the diffusion term, which
forces the ow to respond to both the bed slope and the free surface
slope. The resulting band diagonal system of linear equations are
Fig. 5. Extended plot after Moussa and Bocquillon (1996) showing the parameter values for the Amazon ood wave.
98
spacing of 2 km for the two reaches, 136 for the Solimes reach
and 54 for the Purus reach. These included all parts of the major
channel where it branched around islands but did not include
the oodplain on either side above the low water level. Equivalent
ow area, rectangular cross sections were also derived from the
irregular cross sections for use in LISFLOOD-FP and HEC-RAS.
Two Mannings friction parameters were used, one for the Solimes channel and one for the Purus channel. Fig. 7 shows two
example cross sections from Fig. 6 in more detail. Fig. 7 also shows
the derived equivalent ow area rectangular channel and a background SRTM DEM to illustrate the oodplain topography on either
side of the channel.
Three boundary conditions were applied to the models. The inows for the two channels were extracted from gauged ow data
from Itapeu, 38 km upstream of the model boundary on the Solimes, and Arum, 85 km upstream of the model boundary on the
Purus. For the downstream boundary, a stage hydrograph was derived from Manacapur gauging station stage data. Manacapur is
12 km upstream of the model boundary, so the data were adjusted
using the water slope calculated on a daily basis from the water level difference between the Anam and Manacapur gauging
stations.
Model runs were carried out for the same period used in Wilson
et al. (2007), 1 June 199531 March 1997. This allowed for direct
comparison as well as allowing use of the same calibration and validation data. All models were run with a 12 h time step. Tests
showed results were insensitive to a time step below 24 h, the implicit solvers allowing a large Courant number (Cr 20).
Model testing
Friction parameters were calibrated for the two channels by
varying Mannings n from 0.02 to 0.04 using 0.001 increments for
each channel independently (a total of 441 simulations). Goodness
of t was assessed by comparing model results to water surface
elevation data from four gauges internal to the model domain.
Fig. 6. Amazon channel model schematic. Main gure shows Solimes and Purus reaches with channel cross sections and gauging station locations. Gauged ow data from 1
June 1995 to 31 March 1997 used for (a) Solimes river boundary (Itapeu), (b) Purus river boundary (Arum) and (c) downstream river elevation boundary derived from
Manacapur gauged stage and river slope. Locations of example cross sections illustrated in Fig. 7 are also shown at 36 and 60 km from upstream boundary.
99
(a)
(b)
Fig. 7. Example cross sections, shown facing downstream, extracted from bathymetry data (a) a wide relatively shallow cross section (36 km from upstream boundary) and
(b) a narrow deep cross section (60 km from boundary). Both the irregular cross section and equivalent area rectangular cross section are shown. The background SRTM DEM
(with vegetation artefacts removed using the method described in Wilson et al. (2007) was not used in the channel model and is shown only to illustrate the oodplain either
side of the channel. The grey shaded area is a composite of the SRTM and bathymetry. High and low water elevations are also shown, and denoted HW and LW, respectively.
tions and isolate the effects of each element of the study. The predicted water elevation for each cross section, for the full model
time series on a daily timestep, were compared against the predicted water elevations from the irregular HEC-RAS model water
elevations and an overall RMSE calculated.
In the rst test a varying rectangular channel was compared
with the varying irregular channel. Both model runs were undertaken with HEC-RAS using a full Saint Venant representation of
the ood wave. The second test used the LISFLOOD-FP diffusive
solver with a varying rectangular channel, allowing an assessment
of the impact of simplifying ow representation to a diffusive ood
wave approximation. The third, fourth and fth tests were carried
out with LISFLOOD-FP and test even simpler channel geometries;
one with mean channel width but variable depth, one with mean
channel depth but variable width and the last and simplest representation with mean channel width and depth. The sixth and nal
test uses the mean channel width and depth geometry and the
kinematic solver to test the impact of the channel ow and geometry assumptions made in our previous study (Wilson et al., 2007).
The varying rectangular channel and the mean width and depth
channel are equivalent in ow area to the fully irregular channel.
The mean depth channel and mean width channel do not have
the same ow area and are primarily used to test the effects of estimating only one of these two parameters.
Results and discussion
LISFLOOD-FP model calibration
Calibration of the hydraulic model was straightforward, with
the selection of one model which best ts the observed data from
the matrix of runs obtained by varying the Solimes and Purus
Mannings n values. In this case the best t is determined by the
minimum RMSE for the four gauging stations internal to the model
domain. The LISFLOOD-FP, rectangular cross section, diffusive solver model with the best t had a Mannings n of 0.032 for the Solimes and 0.034 for the Purus. These values are what would be
expected for a large, sand bed river of low sinuosity such as the
100
(a)
(b)
(c)
(d)
Fig. 8. Calibrated model results (dashed line) compared to the four gauging stations (solid line) internal to the model. Solid line is observed data and dotted is model output.
101
(b)
(a)
(c)
Fig. 9. Altimetry water level (circles) compared to predicted water level (crosses) for (a) Solimes, (b) Purus altimetry data locations, with corresponding RMSE for the full
simulation, and (c) altimetry water level versus model water level for all available data and the overall RMSE.
(a)
(b)
(c)
(d)
Fig. 10. Modelled slopes (dashed line) compared to the gauged slopes (solid line), for (a) the Solimes, (b) the Purus reaches. Plots of gauged stage at (c) Codajs and (d) Beruri
are shown below the slope plots for discussion purposes.
the conuence plain which exists between the two rivers, as noted
by Alsdorf et al. (2007a).This acts to reduce the hydraulic slopes in
the Purus. As with the Solimes, the lack of oodplain in the model
means that the simulation fails to reproduce the water surface
slope well at the times when oodplain storage is important. These
fairly large and rapid seasonal changes in the observed water surface slope have been highlighted before by Birkett et al. (2002) and
may be particularly important for studies that use remote sensing
to determine ow from water elevations and slopes, such as the
proposed SWOT mission (Alsdorf et al., 2007b).
102
Table 3
Test models summary and RMSE results.
Test no.
Model code
Process representation
Channel Geometry
Control
1
2
3
4
5
6
HEC-RAS
HEC-RAS
LISFLOOD-FP
LISFLOOD-FP
LISFLOOD-FP
LISFLOOD-FP
LISFLOOD-FP
Saint Venant
Saint Venant
Diffusion wave
Diffusion wave
Diffusion wave
Diffusion wave
Kinematic wave
0.126
0.151
0.414
0.431
0.530
1.319
0.241
0.235
0.565
0.504
0.381
2.941
The 16 model runs done to test the sensitivity of the model results to the downstream boundary condition show that the whole
of both reaches are affected by the downstream boundary, at both
high and low water. For example, decreasing the boundary by 1 m
results in a decrease in water elevation at the upstream end of the
Solimes reach of 0.27 m for high water and 0.05 m for low water.
For the Purus, upstream water levels decrease by 0.40 m and
0.20 m, respectively. Increasing the boundary by 1 m results in
an increase in water elevation at the upstream end of the Solimes
reach of 0.31 m for high water and 0.08 m for low water. For the
(a)
(b)
Fig. 11. (a) Solimes longitudinal prole plot of high and low water results from fully irregular channel model and rectangular channel model, together with both channel
inverts (lled lighter grey is rectangular bed and lled darker grey is fully irregular) and gauged elevations; squares for high water and circles for low water and (b) Purus
prole.
103
Fig. 12. Solimes longitudinal prole plot of high and low water results of the irregular channel model, mean width, mean depth and mean width and depth channel models.
The rectangular channel bed (grey lled) is shown as well as the mean depth bed (dot dash line) and gauged elevations; squares for high water and circles for low water.
with a downstream stage increased by 2 m. This could be explained by the lack of storage in the model resulting in overly steep
slopes, thus raising the downstream stage reduces the slope, giving
a better t. It does imply that compound parameter sensitivity
tests would be worth exploring in the extended model that incorporates the 2D oodplain.
For applications of the model in conditions where downstream
water elevations are not known, for example in climate change
simulations, it is possible to use a stage discharge relationship as
the downstream boundary. This method has not been tested
explicitly in this study, but if used would need to take into account
the hysteresis in the stage discharge relationships for the Amazon
River due to backwater effects noted by Meade et al. (1991).
Diffusive approximation and channel detail tests
A total of seven different models were run to examine the impact of using the diffusive wave approximation, a rectangular
channel and other channel simplications. The same Mannings n
parameter values, derived in the earlier LISFLOOD-FP calibration,
were used for all models The water elevation results at all 190
cross sections for these runs were compared to the results of the
full Saint Venant, fully irregular HEC-RAS channel model. This control model allowed a relative comparison, isolating the effect of
each test. A summary of the models and water elevation RMSE results are presented in Table 3.
The rst test explores the effect of simplifying the channel cross
sections from fully irregular sections to rectangular sections of the
same ow area. The RMSE results show that the impact on the results of this simplication is small and demonstrates that a wide
rectangular channel representation is a reasonable approximation
in this case. Longitudinal plots of the high and low water results
for the two models are shown in Fig. 11 together with the bed inverts for the irregular and rectangular channels.
The second test exploring the effect of using the diffusive wave
approximation instead of the full Saint Venant solution also shows
that the impact is minimal. Given that the results of the ood wave
characterisation show that a diffusive wave approximation should
be valid, this is expected, but does provide further support for our
approach.
The three bathymetry tests use simplications to the geometry
of the channel to explore how important the information content is
to the results. Prole results for the three models at high and low
water are presented in Fig. 12. All three model results give RMSE
values in the order of 0.5 m. This is true even for the simplest representation, which is essentially one rectangular cross section per
channel with a slope. This demonstrates that compared to the
amplitude of the Amazon ood wave, results are relatively insensitive to the bathymetry information content of the channel model.
This means that for studies of the central Amazon channel such
as the virtual mission for the SWOT mission (Alsdorf et al.,
2007b) even relatively crude assumptions regarding the bathymetry will be valid as long as they can approximate well the true
mean width and depth for a reach. Remote measurements of water
elevations and hence derived slope and discharge will implicitly
take into account backwater effects through the slope.
The nal test uses the simplest channel geometry as well as the
kinematic solver and shows the signicant error, of the order of 1
3 m, introduced by ignoring the diffusive term in the Saint Venant
equations. Overall these results show that we need at least a diffusive wave representation to match observed data and include
backwater effects, but that it is not necessary to represent the
oodplain for short reaches of around 100200 km and still obtain
a good t to observed data. They also show that a rectangular channel approximation is valid and it is possible to get a reasonable t
with an error in the region of 0.5 m on a 1112 m ood wave
amplitude using simple cross section and slope models. Finally,
in terms of modelling overall hydraulics at this scale, the result
that using a mean depth does not lead to large errors implies that
the model is relatively insensitive to the bed forms. This means
that it is likely that the changing bed conditions can be ignored
for model runs of less than decadal timescales. For longer timescales it may be necessary to assess the importance of channel
migration, although an assumption that the channel retains a similar ow capacity may still be valid in this case.
Conclusions
A large quantity of original and up-to-date bathymetric information for the central Amazon River channel has been collected
for a 430 km reach of the Solimes River and a 145 km reach of
the Purus River. These data show the complexity of the river bed
structure and provides an unprecedented rst order data set for
hydraulic analysis. For the Solimes reach, the channel width
104
varies from 1620 m to 5624 m with a mean of 3711 m and the minimum bed elevation, varies from 26.45 m to 8.03 m with a mean
of 8.44 m (Kosuth et al. (2006) datum). At high water, channel
depths range from 20 m to 52 m. For the Purus reach the channel
width varies from 600 m to 1678 m with a mean of 1114 m and
the minimum bed elevation varies from 9.78 m to 9.48 m with
a mean of 2.08 m. At high water, channel depths range from
16 m to 35 m.
Our investigations show that the Amazon ood wave is subcritical and diffusive in character. For the whole Solimes reach during
the period of 1996/1997, mean model ow depth varies between
21.9 m at low water and 33.5 m at high water, giving a ood wave
amplitude of 11.6 m. Mean bathymetric bed slopes are 5.0 cm/km
with a mean gauged water surface slope of 2.8 cm/km. The ow is
deep and relatively slow moving, resulting in very low Froude
numbers of the order of 0.05.
Due to very shallow slopes, backwater conditions control significant reach lengths in the central Amazon and these backwater
conditions are present for low as well as high water states. The
modelled lower reaches of the Purus tributary also show backwater effects from the conuence with the Solimes main-stem. To
accurately predict water elevations using hydraulic models requires that these backwater effects are represented.
We have demonstrated a successful implementation of a diffusive channel solver to the LISFLOOD-FP model and its application
to a channel only model of the central Amazon. Calibration of the
hydrodynamic hydraulic model gives a Mannings n of 0.032 for
the Solimes and 0.034 for the Purus. Calibrated RMSE for the
water elevations at the four gauging stations internal to the model
are all less than 1 m for low and high water results. Validation
using water elevation altimetry data at one location on each river
gives RMSEs under 1.5 m, and comparing the model to gauged
slopes, gives a maximum RMSE of 0.62 cm/km. This successful
application to the Amazon builds on previous work by Wilson
et al. (2007) and will allow further development of an integrated
1D channel and 2D oodplain model in order to provide quantitative inputs to biochemical and geomorphic studies requiring detailed hydrodynamic information.
Experimentation with the physical process representation of
the channel ow as well as bathymetric information content of
the channel shows that when compared to the amplitude of the
Amazon ood wave, water levels are relatively insensitive to the
bathymetry information content of the channel model. This means
that for studies of the central Amazon channel such as virtual missions for the proposed SWOT satellite mission, even relatively
crude assumptions regarding the bathymetry will be valid as long
as the mean cross sectional area can be reasonably well approximated. Remote measurement of water elevations and hence a derived slope and discharge will also implicitly take into account
backwater effects through the water surface slope. These conclusions may well be applicable to other large rivers, where similar
hydraulic conditions prevail, specically the shallow sloped lower
reaches. In addition, SWOT satellite mission measurements will
also provide valuable calibration and validation data for future
hydraulic modelling.
For the application of hydraulic modelling methods to these
reaches of the central Amazon, it is has been demonstrated that
it is necessary to include at least the diffusion term in the channel
model. The error on predicted water elevation introduced by using
a wide rectangular channel is in the order of 0.100.15 m and by
ignoring the acceleration and advection terms we introduce a further error in water elevation of the order of 0.020.03 m. Both
these errors are very small in comparison to the mean annual ood
wave amplitude of 1112 m. The application of appropriate boundary conditions has also been demonstrated as essential in order to
incorporate the backwater effects present along these reaches.
105